The report analyses the intersection of cloud infrastructure and AI services offered by leading providers—Amazon Web Services (AWS), Google Cloud Platform (GCP) and Microsoft Azure
Cloud environments leveraging artificial intelligence (AI) services are highly susceptible to cyber risks, according to new research from Tenable®, the exposure management company. The findings, published in Tenable’s latest Cloud AI Risk Report 2025, reveal that nearly 70 per cent of cloud workloads using AI contain unresolved vulnerabilities, posing serious security threats to sensitive AI data and models.
The report analyses the intersection of cloud infrastructure and AI services offered by leading providers—Amazon Web Services (AWS), Google Cloud Platform (GCP) and Microsoft Azure. While both cloud and AI technologies have become essential for modern businesses, Tenable warns that their combination introduces layers of complexity, often creating overlooked security gaps.
One of the key concerns highlighted in the research is the presence of unremediated vulnerabilities in the majority of cloud AI workloads. Notably, the study found that CVE-2023-38545, a critical vulnerability in the popular curl library, was present in 30 per cent of these workloads. This vulnerability could potentially allow attackers to manipulate data flows within cloud-hosted AI environments.
Further, the report draws attention to what Tenable terms “Jenga-style” misconfigurations—a concept describing how cloud providers tend to build one service on top of another, often carrying forward risky default settings. Specifically, 77 per cent of organisations using Google Vertex AI Notebooks were found to be using an overprivileged default Compute Engine service account, exposing their services to unnecessary risk.
Amazon’s AI offerings were also flagged. According to the report, 91 per cent of Amazon SageMaker users have at least one notebook instance configured to grant root access by default. If compromised, these instances could allow unauthorised users to modify or access sensitive files. Additionally, 14 per cent of organisations using Amazon Bedrock have not blocked public access to at least one AI training bucket, while 5 per cent have overly permissive buckets, leaving AI training data vulnerable to data poisoning attacks that could skew model outcomes.
“When we talk about AI usage in the cloud, more than sensitive data is on the line. If a threat actor manipulates the data or AI model, there can be catastrophic long-term consequences, such as compromised data integrity, compromised security of critical systems and degradation of customer trust,” said Liat Hayun, Vice President of Research and Product Management, Cloud Security, at Tenable. “Cloud security measures must evolve to meet the new challenges of AI and find the delicate balance between protecting against complex attacks on AI data and enabling organisations to achieve responsible AI innovation.”
The report underscores the importance of strengthening cloud AI security practices to mitigate potential breaches and safeguard data integrity. As businesses increasingly turn to cloud-based AI services, Tenable urges them to regularly audit their configurations, patch known vulnerabilities and implement strict access controls to avoid falling victim to these avoidable risks.
